Nervous System Targets of RNA Editing Identified by Comparative Genomics

See allHide authors and affiliations

Science  08 Aug 2003:
Vol. 301, Issue 5634, pp. 832-836
DOI: 10.1126/science.1086763


An unknown number of precursor messenger RNAs undergo genetic recoding by modification of adenosine to inosine, a reaction catalyzed by the adenosine deaminases acting on RNA (ADARs). Discovery of these edited transcripts has always been serendipitous. Using comparative genomics, we identified a phylogenetic signature of RNA editing. We report the identification and experimental verification of 16 previously unknown ADAR target genes in the fruit fly Drosophila and one in humans—more than the sum total previously reported. All of these genes are involved in rapid electrical and chemical neurotransmission, and many of the edited sites recode conserved and functionally important amino acids. These results point to a pivotal role for RNA editing in nervous system function.

Chance discoveries of genetic recoding by modification of adenosine to inosine (A-to-I) in precursor mRNAs (pre-mRNAs) have been reported for more than a decade. The modification reaction, a hydrolytic deamination, is carried out by the adenosine deaminases acting on RNA (ADARs) (1, 2). Inosine in mRNA is recognized as guanosine (G) by cellular machineries, including the ribosome in the course of protein synthesis. The archetypal target of specific A-to-I RNA editing is the transcript of the mammalian ionotropic glutamate receptor gene, GluR-B. At the GluR-B (Q/R) editing site, an encoded glutamine (Q) codon (CAG) is enzymatically modified to CIG in mRNA, which is then decoded as an arginine (R) codon (CGG). The conductance properties of the edited channel differ from those of the genomically encoded GluR-B protein (3). Thus, RNA editing expands the protein repertoire of genes, a function that is important for animal life given that Caenorhabditis elegans, Drosophila, and mouse mutants lacking ADAR enzymes display predominantly neurological phenotypes (47).

One factor frustrating ADAR site discovery is the lack of a signature sequence motif. The pre-mRNA substrate required by an ADAR enzyme is usually an imperfect duplex RNA formed by base-pairing between the exon that contains the adenosine to be edited and an intronic noncoding element, called the editing site complementary sequence (ECS) (8). An ECS can be several hundred to several thousand nucleotides upstream or downstream of the edited adenosine (810). In short, any given ADAR editing site comprises the adenosine(s) to be modified within a stretch of gene-specific coding sequence, plus a pairing partner whose presence is inferred and whose location must be determined experimentally.

To investigate the substrate requirements of ADAR enzymes, we undertook a phylogenetic study of an editing site of the para Na+ channel gene, which is edited in numerous Drosophila species (11). We hypothesized that if RNA editing of a particular site were conserved between species, then intronic cis-elements required for editing site/ECS duplex formation would be conserved as well. Unexpectedly, our approach revealed highly conserved exonic sequences neighboring sites of ADAR modification. Mutations, including synonymous changes, were virtually absent near sites of ADAR modification in 18 Drosophila species tested (fig. S1) (12). Sequences within the same exon, distal to the region of ADAR modification, acquired substitutions at a much higher rate. Assuming that RNA editing confers a selective advantage upon organisms, we interpret this highly conserved region surrounding an editing site as having arisen from a selective constraint against any mutation near a site of ADAR modification. Such mutations would perturb the secondary structure required for editing and have been shown experimentally to disrupt or diminish editing (8, 9, 13). A relaxation of this constraint would be expected in flanking coding sequences not involved in editing.

Using this high degree of sequence identity between species as a potential signature of ADAR editing sites, we searched for new ADAR targets. Several groups of genes in Drosophila melanogaster were compared with their orthologs in Drosophila pseudoobscura. Our choice of genes in this study was directed, in part, by the neurological phenotype of dADAR null mutants (6). We examined 914 genes: those annotated as ion channels (n = 135), G protein–coupled receptors (GPCRs, n = 178), proteins involved in synaptic transmission (n = 102), and transcription factors (n = 499). Of these, 41 genes contained regions within coding sequences that displayed unusually high sequence conservation compared with surrounding sequences (Fig. 1, A and B).

Fig. 1.

Prediction and experimental confirmation of an RNA editing site in Drosophila. (A) Scale representation of the D. melanogaster DSCI (CG9071) Na+-channel transcription unit in the region of the RNA editing site. Boxes represent exons and the line represents introns. Red box indicates the exon containing the editing site. (B) Bar graph showing percent sequence identity for the exons of DSCI transcript CG9071-RA. Exon numbering corresponds with CG accession number. Red bars indicate the edited exon, blue bars represent the flanking exons shown in (A), and black bars are all other exons of the transcript. (C) Sequence analysis of DSCI RT-PCR amplification products from whole-fly RNA. Shown are electropherograms of products generated from dADAR+ (top), dADAR (middle), and D. pseudoobscura flies (bottom). Sequences are labeled above the electropherogram in codon triplets with editing sites indicated by mixed A/G signals. Amino acid sequences for each codon are shown above the nucleotide sequence with the amino acid change shown for the editing site by an arrow. (D) An additional editing site seen in D. pseudoobscura. Electropherogram from D. pseudoobscura RT-PCR products (top) and D. pseudoobscura genomic DNA amplification products (middle). No editing was seen in D. melanogaster dADAR+ flies (bottom). (E) Protein sequence alignment of Na+-channel sequences compared to DSCI. Red shading indicates identity with DSCI. Blue shading indicates conservative changes. Editing sites are highlighted in yellow.

To determine whether any of these genes are subject to editing, we performed reverse transcription–polymerase chain reaction (RT-PCR) followed by sequence analysis of amplification products for the 41 candidate genes in both wild-type and dADAR-deficient genetic backgrounds (Fig. 1C). Electropherograms from 16 of the 41 genes showed one or more sites with mixed A/G peaks in a wild-type background, indicative of a mixture of edited and unedited transcripts. In the same analysis involving dADAR animals, all sites yielded pure A signal in sequenced products.

Our method of discovering new ADAR targets introduces no a priori assumptions and makes no predictions as to the identity or number of targeted adenosines or the level of editing for a given site. For example, the D. pseudoobscura Drosophila Sodium Channel I (DSCI) gene contains the same editing site identified in the D. melanogaster gene, although the level of editing is higher for the former (Fig. 1C). The same exon of the D. pseudoobscura gene contained an additional editing site that was absent from the D. melanogaster gene, although both species encode identical exonic sequences (155/155 nucleotides) (Fig. 1D). We presume that species-specific variation of the ECS element for this editing site is responsible for the different pattern of ADAR modification observed between species.

All of the ADAR targets we identified by our method function in rapid electrical and chemical neurotransmission (Table 1 and table S1). The majority of ADAR targets identified were voltage-gated ion channels (VGICs) or ligand-gated ion channels (LGICs). Most VGICs are tetrameric integral membrane proteins consisting of a central ion-specific conduction pathway (S5-pore-S6) surrounded by portions of the protein involved in voltage-dependent gating (S1 to S4) (Fig. 2, A and B). The gating of VGICs upon membrane depolarization results in channel opening and flow of ionic currents. Frequently, current flow is ended by another gated process, channel inactivation. The firing properties of a given neuron are determined largely by the gating characteristics of the spectrum of VGICs expressed there. We found both voltage-gated Na+ and Ca2+ channels in our screen, channels which serve to depolarize neuronal membranes. The editing sites in these channels occur in transmembrane (TM) domains or in other functionally relevant portions of the protein involved in channel gating or inactivation (Fig. 2A). For instance, editing of the DSCI Na+ channel occurs within the cytoplasmic linker between HD-III and HD-IV, termed “the inactivation gate.” The editing site is in the functionally critical IFM motif [in this case, (M→ V)FL] (Fig. 1E). Studies have shown that mutational substitution or deletion of these residues alters the inactivation kinetics or renders the channels non-inactivating (14).

Fig. 2.

Nervous system targets of RNA editing in Drosophila. (A) The four internally homologous six–transmembrane domain repeats of Na+, Ca2+ channels, which are evolutionarily related to K+ channels. Homology domains (HDs) I to IV are indicated. The transmembrane domains S1 to S6 are labeled in HDII. Plus symbols (+) indicate the S4 voltage sensors. RNA editing sites for DSCI (red), T-type Ca2+ channel (yellow), and L-type Ca2+ channel (blue) are shown as spheres. (B) The basic K+-channel subunit. Labels are as shown in (A). RNA editing sites for Sh (red), eag (yellow), and slo (blue) are shown as spheres. The different COOH domains of the respective channels are labeled. (C) The editing sites of Sh, eag, and slo, which fall into the S5-pore-S6 region shown on a ribbon diagram based on the crystal structure of the KcsA channel. The amino acid side chains analogous to the editing-site amino acid positions are colored as in (B). (Left) A top-down representation looking at the outer vestibule of the KcsA channel. (Right) A side view with two opposing subunits removed. K+ ions in the crystal structure are shown in gray. (D) Crystal structure of rat synaptotagmin I C2B domain. Ribbon structure of the backbone is shown with the editing-site residues highlighted in yellow and labeled (A) to (D). Red lysine-residue side chains of the KKKK (fig. S2) motif are also shown. Ca2+ ions are shown as spheres. Functional domains of syt I are labeled. (E) RNA editing sites are shown on a schematic of a model LGIC. Transmembrane domains TM1 to TM4 are labeled 1 to 4. Editing sites in Rdl (red), Dα5 (dark blue), ARD (light blue), and SBD (yellow) are shown as spheres. The hydroxyl groups of the pore-lining S/T residues of TM2 are shown as –OH groups.

Table 1.

Gene targets of A-to-I RNA editing in Drosophila identified by comparative genomics. Confirmed RNA editing sites in a screen of 914 genes in Drosophila fall into three classes of gene products involved in fast electrical and chemical neurotransmission.

Name Protein function Number of sites Functional consequenceView inline Ref.
Voltage-gated ion channels
DSCI Na+ channel 1 +++ (View inline)
Ca-alpha1T Ca2+ channel 1 +
DmCa1D Ca2+ channel 5 ++
α2δ Ca2+ channel accessory subunit 3 +
Shaker (Sh) K+ channel 6 +++ (View inline, View inline)
ether-a-go-go (eag) K+ channel 6 +++ (View inline)
slowpoke (slo) K+ channel 2 +
Synaptic release machinery
Synaptotagmin (syt) Ca2+ sensor 4 +++ (View inline)
Dunc-13 SNARE binding 1 ++
Stoned B (stnB) ? 1 +
complexin (cpx) SNARE protein 3 ?
lap Adaptor protein 1 ?
Ligand-gated ion channels
Dα5 nAChRα subunit 7 +++ (View inline)
ARD nAChRβ subunit 4 ?
SBD nAChRβ subunit 2 +++ (View inline)
Resistance to dieldrin (Rdl) GABA-receptor 6 +++ (View inline)
  • View inline* Evidence supporting functional importance of RNA editing for each target gene is indicated by the following: (+++) structure/function studies and sequence invariance; (++) sequence invariance; (+) sequence conserved; (?) unknown.

  • Potassium channels in the nervous system serve many roles, such as maintenance of resting membrane potential, membrane repolarization, and tuning the firing properties of a given neuron. We identified multiple editing sites in three K+ channel genes: Shaker(Sh), ether-a-go-go(eag), and slowpoke(slo) (Fig. 2B). The design of the conducting pore of K+ channels is ancient, as can be seen by the similarity in crystal structures of the non—voltage-gated KcsA and voltage-gated KvAP channel pores, which share sequence homology with all voltage-gated ion channels (15, 16). slo site A, eag site A, and Sh site C can all be placed into the crystal structure of the KcsA channel and lie in the extracellular vestibule of the channel or within the conducting pore (Fig. 2C). Other editing sites in these K+ channels are found in highly conserved regions involved in channel gating and modulation.

    Electrical impulses in neurons terminate with an increase in intracellular Ca2+ and neurotransmitter release at a synapse. Unexpectedly, we identified several components of the synaptic release machinery that undergo RNA editing. The best-studied of the identified targets is synaptotagmin (syt), whose C2 domains serve as the Ca2+ sensor for neurotransmitter vesicle fusion (17, 18). Editing at several positions in the syt C2B domain was observed in regions essential for syt self-association, binding to Ca2+ channels, and association with accessory proteins involved in vesicle reuptake (Fig. 2D) (19). Editing was also seen for other proteins of the core-complex synaptic release machinery in C2 domains or other interaction surfaces within the core-complex (Table 1).

    Rapid responses of a neuron to vesicle fusion and neurotransmitter release into the synaptic cleft are accomplished through the action of LGICs that either excite or inhibit postsynaptic neuronal output. We found both kinds of LGICs in our screen: nicotinic acetylcholine receptor (nAChR) α and β subunits and a γ-aminobutyric acid (GABA) receptor, Resistance to dieldrin (Rdl) (Table 1 and Fig. 2E). Editing of these receptors in Drosophila occurs in several conserved regions including the ligand-binding domain, TM domains involved in gating and agonist sensitivity and, importantly, the lining of the conducting pore, TM2 (Fig. 2E). Editing sites found in TM2 in these channels are in hydrophilic or charged residues that have been shown by mutagenesis studies to affect ion selectivity and channel conductance (20, 21).

    We applied our screening strategy to mammalian Shaker genes (Kv1.1-5). Orthologs of the Kv1.1 gene in mouse, rat, and human showed a region of unusually high sequence conservation not seen in the other family members. Subsequent analysis revealed RNA editing of a site within this conserved region in all three species (fig. S3). RNA editing of the human Kv1.1 ortholog (KCNA1) was spatially regulated (Fig. 3A). Levels of editing ranged from 17% in the caudate nucleus to 77% in the medulla. We also observed high levels of editing in spinal cord RNA (68%). A comparison of Kv1.1 editing in different regions of mouse and rat brain revealed spatial differences consistent with those seen in humans (Fig. 3A). Thus, spatial regulation of RNA editing for Kv1.1 genes is evolutionarily conserved.

    Fig. 3.

    RNA editing of the human KCNA1 gene. (A) Quantitation of editing levels from various brain and nervous system regions from human (blue), mouse (red), and rat (yellow). (B) Protein alignment of the pore-lining S6 transmembrane domain from various Kv channels along with the bacterial KcsA channel. Invariant residues are indicated by red shading; blue indicates greater than 75% identity with human KCNA1. The position of the editing site is highlighted in yellow. The arrow indicates the position of Drosophila Sh site C. The asterisk indicates the position of Drosophila eag site A. (C) Proximity of the I400 residue editing site to the blocking particle. Side view of the KcsA crystal structure with the quaternary amine blocker TBA, which models the inactivation gate (green). The KcsA phenylalanine (F) side chain (yellow) from the analogous position of the Kv1.1 editing site is shown. K+ ions in the crystal structure are shown as pink spheres.

    The location of this editing site within the Kv1.1 protein is especially provocative. The isoleucine (I) to valine (V) change introduced by editing is in a highly conserved position (I400 in human KCNA1) within the pore-lining S6 domain of Shaker-type K+ channels (Fig. 3B). Structural work indicates that the inactivation particle of a Shaker-type channel or its accessory β subunit enters the channel and binds to residues in the pore itself, blocking conduction (22). Hydrophobic residues at the N-terminus of the particle enter the inner vestibule and interact strongly with the I400 residue, as do quaternary amine inhibitors such as tetrabutylammonium (TBA). Directed substitution of this residue by the smaller alanine residue reduces the affinity of the pore for the blocking particle by more than 400-fold, resulting in channels that either fail to inactivate or inactivate incompletely, and that consequently pass more current than the wild-type channel. In the case of Kv1.1 channel editing, the I→ V change also reduces the amino acid side-chain volume at this position that conceivably would have functional consequences.

    The genes identified here as targets of genetic recoding by ADARs all serve prominent roles in rapid signaling in the nervous system. This is despite the fact that our screen included a majority of genes not directly involved in fast electrical or chemical signaling. The preponderance of ADAR-mediated codon reassignment in the nervous system is supported by the neurological phenotypes of animals lacking ADARs. Given the spatially and temporally hierarchical events that occur in neurons on the time scale of milliseconds, we propose that the advantage of regulating VGICs, LGICs, and the synaptic release machinery by RNA editing lies in a finer level of control than discrete genetic change permits. Clearly, different species have chosen different subsets of these targets to edit in order to modify signaling in their nervous systems. Yet, a commonality does exist, in that both fruit flies and mammals edit their Shaker channels.

    Certain dominant neurological disorders such as episodic ataxia-1 (EA-1) and autosomal dominant nocturnal frontal lobe epilepsy (ADNFLE) result from heterozygosity for a missense mutation in some of the analogous positions or regions to editing sites identified in this study (23, 24). The implications of this for human genetic disease are that mutations of the recoding process of RNA editing, particularly when it targets functionally critical residues in proteins, may be a source of variation or disease. Moreover, certain mutations in noncoding regions (for example, the ECS) may affect particular A-to-I RNA editing sites while being far removed from the editing site itself. As the number of human ADAR targets increases, such mutations will need to be considered as another potential source of inherited behavioral differences as well as neurological disorders.

    Supporting Online Material

    Materials and Methods

    Figs. S1 to S18

    Table S1

    References and Notes

    View Abstract

    Navigate This Article